Fuzzy Matching

An Additional Level of License Plate Read Verification

Although license plate recognition algorithms are the most mature video analytics on the market, when using any license plate recognition solution, it is possible that license plates will be misread. To maximize wanted-vehicle identification or automate gated access control with minimal user intervention, Genetec’s AutoVu solution uses fuzzy matching algorithms. Available in both the in-vehicle system, the AutoVu Patroller, as well as in the back-end system within Genetec’s unified security platform, the Security Center, the Fuzzy Matching feature in AutoVu essentially adds another level of verification to ensure that even when a plate may be misread, operators are really getting the best possible matches of every license plate to the database of vehicles of interest.

LPR Technology and Real-life Applications

The result of intensive research and tests conducted over the last 30 years suggests that license plate recognition (LPR) systems are the most mature form of video analytics in the security industry. The performance of LPR systems today has reached a high-accuracy level enabling their use in many demanding applications such as law enforcement, permit and time-limit parking enforcement, access control or city-wide and traffic surveillance.

When using any type of video analytics, there is a trade-off between alerting users of a false alarm and not alerting them of a critical alarm. When looking for wanted vehicles (i.e., stolen vehicles, wanted felons, amber alerts, expired registrations, revoked licenses or scofflaws) with the AutoVu LPR system, law enforcement officers benefit from having a false alarm on a potential match because it decreases the chances of missing a wanted vehicle. Parking enforcement officers want to make sure that each person that owes revenue to the city pays their outstanding fines. When controlling access to a facility, depending on the chosen security level, security managers may want to prioritize automated access control on a restricted list of permitted vehicle versus operator intervention.

What Is Fuzzy Matching?

Fuzzy matching actived, officer alerted of hotlist match

The AutoVu solution not only matches exactly each license plate read to the license plates in the various hotlists, but also indicates any potential matches to license plates with Fuzzy Matching. Fuzzy matching is a mathematical notion that integrates vagueness in a decision system. Unlike a true or false decision, you will get several maybe levels. When applied to LPR in matching algorithms, fuzzy matching gives the ability to match a license plate containing one or more errors in the read, against a hotlist and sends an alert on any potential matches. Errors in the read may be introduced due to uncontrollable environmental factors such as dirt or snow accumulation on the license plate. This fuzzy matching extends the comparison of the license plate to probable results (i.e., visually similar license plates or incomplete license plates). Regardless of whether AutoVu is used in a fixed application as part of the unified security platform, the Security Center, or in a mobile application through the AutoVu Patroller, fuzzy matching is a built-in feature that operators automatically benefit from.

How It All Works

Although today’s LPR solutions deliver high accuracy rates, some license plate characters are visually similar. Depending on the font and the field of view of the camera used, characters may be confused for one another, for example, a “2” and “Z” or, “8” and “B”, or “0”, “O”, “D”, “Q”, we call this optical character recognition(OCR) equivalents.

License plate ABC123 could be read as ABCI23 or A8C1Z3. Without fuzzy matching, a license plate ABC123 in a list of wanted vehicles would not match any of the above interpretations of the actual license plate, hence potentially missing a critical alert. With fuzzy matching, any of the visually similar license plates would be recognized as a potential match and the officer would get an alert. The alert can be validated by verifying the additional information from the hotlist, such as issuing state, VIN, vehicle make or model. Fuzzy matching provides law enforcement officers with critical information which would otherwise go unnoticed.

Fuzzy matching activated, gate opens

Fuzzy matching goes further than compensating for OCR equivalents. It also suggests potential matches on a hotlist for license plates which are misread (i.e., for which there is one misread character or with one too many characters). When reading license plates, some license plates are partially obstructed by hitch, have an illustrated background, or have an additional symbol engraved on the license plate (i.e., a handicapped symbol), which impacts the readability of the plate. What if an ABC123 license plate is read A81Z3? Without fuzzy matching, every time a license plate is not perfectly read, the security operator managing the access to a parking lot would have to manually verify the license plate to confirm if the vehicle is permitted to park in the lot. With fuzzy matching this can be automated to minimize manual intervention.

Customized Objectives and Benefits

Fuzzy matching can be tailored to what you are trying to accomplish. In other words, the threshold for tolerance when cross-referencing a plate against a hotlist can be adjusted to limit the number of false-positive reads. Whether from the back-end system in the Security Center, or directly from the in-vehicle application, the AutoVu Patroller, law enforcement officers can configure the system to match any vehicle to a hotlist while considering OCR equivalents only or OCR equivalents and misread characters. Officers validate if the vehicle found is in fact the vehicle of interest in the hotlist by using the additional information in the hotlist such as issuing state, vehicle make, model, year or VIN. If officers do get an alert for license plate AYC13 on a Ford Mustang 2003 from Washington, when looking for license plate ABC123 on a Ford Focus 2000 from Virginia, then the officer can quickly dismiss the hit and continue patrolling the streets. Security managers can automate access to their parking facilities while keeping a maximum number of unauthorized vehicles out and keeping manual verification to a minimum by authorizing any vehicles with license plates containing OCR equivalents to access their parking facilities.